🔋 Romania’s Grid Has a €28 Million Monthly Problem. The Nuclear Shutdown Just Made It Impossible to Ignore.
Both Cernavodă reactors went simultaneously offline for the first time in history. Prices hit 140 EUR/MWh. Romania became Europe’s most expensive electricity market in under 48 hours.
Romania sold electricity in April 2026 at 50 EUR/MWh.
Then bought it back at 250.
Not across different weeks. Not during a geopolitical crisis. On the same day — midday surplus exported, evening peak imported — four hours apart, five times the price. The result: a net monthly outflow of approximately 28 million EUR. Not because Romania lacks generation capacity. Because the capacity it has generates when the sun shines, and the demand peaks when it doesn’t.
The nuclear outage that began on May 4 didn’t create that problem. It widened the gap until even people who weren’t watching Romanian energy markets noticed.
This piece is the full account.
⚛️ What Happened at Cernavodă — and Why the Timing Matters
The insulator that started it
At 19:35 on May 4, 2026, an insulator on a power evacuation transformer at Cernavodă Unit 2 degraded under operating conditions. The plant’s protective systems detected the anomaly and triggered automatic disconnection. Standard sequence. Nuclearelectrica confirmed the cause, announced transformer replacement, and set a restart target of June 1, 2026.
700 MW of constant baseload generation: offline.
Unit 1’s scheduled maintenance window was approved independently, well in advance. It disconnected on May 10 at 11:00 AM, per plan.
The timing overlap was unlucky. Neither outage was mismanaged. Both followed their own engineering logic. The problem is that Romania’s grid had never absorbed both simultaneously — because in thirty years of operation, it hadn’t happened. System planning had not priced that scenario into its architecture.
What followed was the first simultaneous shutdown of both CANDU reactors in Romanian nuclear history.
The reactors themselves — why this matters beyond the headlines
Cernavodă is not just Romania’s only nuclear plant. It is the only CANDU-design reactor in Southeast Europe — a technology renowned for its operational flexibility and exceptionally high capacity factors. Together, the two units contribute approximately 18–20% of Romania’s annual electricity generation. Not as an average across weather conditions, but as a constant. Available on demand. 24 hours a day. Every day.
That constancy is what the rest of the grid is built around. Romania can aggressively expand solar capacity precisely because 1,400 MW of reliable baseload is always there to balance it. Remove that anchor, and every percentage point of renewable penetration becomes harder to manage.
The structural picture behind the three-week crisis
Unit 2 returns around June 1. The immediate pressure eases.
Unit 1 does not. In December 2026, it enters a planned refurbishment programme worth €1.85 billion — one of the largest energy infrastructure investments in Romania’s post-communist history. The refurbishment extends the unit’s operational life by decades. It is the right investment. It also means Romania operates on a single reactor — 700 MW of baseload — from December 2026 through December 2028.
Two years. What happened in May 2026 was a preview. Starting in December 2026, it becomes the structural baseline.
📊 The Grid Mathematics of Losing 23% of Dispatchable Capacity
Installed capacity is not the number that matters
Romania’s total installed generation capacity sits at approximately 20,000 MW. That figure is almost meaningless for understanding what happened on May 10.
The number that matters is dispatchable capacity — generation that can be called when demand peaks, regardless of weather. Wind produces when the wind blows. Solar produces when the sun shines. Hydro depends on reservoir levels. Only nuclear and thermal generation are fully controllable. Of Romania’s dispatchable fleet, Cernavodă represents approximately 23% of peak demand.
That 23% isn’t just capacity removed from the stack. It’s the reliable slice that everything else is planned around. Lose it, and the grid’s weather-dependency increases sharply — and immediately.
What 2,200 MW of imports actually means
On the evening of May 10 at 21:35, Romania was importing approximately 2,200 MW from neighbouring grids.
Two things are worth dwelling on.
First: that figure exceeds Romania’s total hydroelectric output at that hour. Romania’s rivers contribute significant generation capacity — and on May 10, the country was leaning harder on its interconnections than on its own water. Second: when Romania imports from the European interconnected grid, it doesn’t pay its own marginal cost of generation. It pays the marginal clearing price of the grid it’s drawing from.
If Romania’s domestic cost is 80 EUR/MWh and the interconnected grid is clearing at 140 EUR/MWh, Romania pays 140 EUR/MWh. That’s not a failure. That’s how interconnected markets work. But at 2,200 MW scale, on a night when domestic shortfall is structural rather than marginal, the arithmetic becomes expensive very quickly.
💸 How the OPCOM Day-Ahead Market Broke — and Why It Was Predictable
The clearing mechanism
OPCOM’s day-ahead market (DAM) is where Romanian electricity is priced and traded — approximately 75–80% of all traded volume flows through it. Sellers submit generation offers. Buyers submit demand bids. The market clears hourly, setting a unique price for each hour of the following day.
The critical mechanic: the last marginal unit to clear in any hour sets the price for that entire hour. When domestic supply drops and imports must fill the gap, import costs set the marginal clearing price. Every Romanian generator gets paid that import price too — a windfall for domestic producers, a bill for Romanian consumers and grid operators.
On May 11, 2026, Romania’s average DAM price reached approximately 730 lei/MWh — roughly 140 EUR/MWh. Highest in Europe that day.
April 2026 weighted average: 514.65 lei/MWh — approximately 99 EUR/MWh.
42% jump. 48 hours.
Why 140 EUR/MWh is not the story — the 200 EUR/MWh spread is
The 140 EUR/MWh headline number has attracted most of the commentary. It shouldn’t. It’s a symptom.
The structural story is in the spread.
The spread on May 10 was not an anomaly. It was the same arbitrage gap that had been running throughout April — the nuclear outage simply made the DAM average visible in a way that generated coverage.
What the table shows: Romania had a 200 EUR/MWh intraday arbitrage spread throughout April 2026. Normal baseline spreads run 30–40 EUR/MWh. This was five to seven times the baseline. And it cost approximately 28 million EUR in net monthly outflow — exporting at 50, importing at 250, all month long.
Romanian energy analysts at AEI described this as “rezultatul unei proaste planificări” — the result of poor planning. I’d frame it differently. It’s the result of a grid designed for one generation mix operating in a fundamentally different one. The solar capacity was correctly built. The storage capacity to monetise it was not.
🔋 The BESS Revenue Case — Built From the Actual Numbers
What May 10 would have paid
Let me build this precisely.
A 150 MW / 300 MWh BESS asset connected to Romania’s transmission grid on May 10 would have run a standard DAM arbitrage cycle.
Charge during midday solar surplus at approximately 50 EUR/MWh. Discharge during the evening peak at approximately 250 EUR/MWh. Spread: 200 EUR/MWh. Volume: 300 MWh.
DAM arbitrage revenue: 60,000 EUR. One asset. One day.
That’s not the full picture.
The frequency services layer
When 1,400 MW of synchronous baseload generation disappears, grid frequency deviates from 50 Hz. Transelectrica must restore it using FCR (Frequency Containment Reserve) and aFRR (automatic Frequency Restoration Reserve) assets — both compensated at ancillary market clearing prices. Under normal grid conditions: 60–70 EUR/MWh equivalent. Under grid stress, they spike above that range.
A BESS pre-qualified for FCR and aFRR would have been actively dispatched during the frequency excursions caused by the dual nuclear outage. Ancillary revenue on top of the DAM arbitrage cycle. This is the part of the BESS commercial case that rarely gets modelled properly by developers coming to the asset class for the first time: the revenue stack is not just correlated with grid stress events. It is positively driven by them. The asset earns most in the exact conditions where the grid needs it most.
The balancing market — the third layer
Transelectrica’s real-time balancing mechanism corrects imbalances between scheduled and actual generation. Prices are the most volatile in the stack — typically 80–120 EUR/MWh under normal conditions, with significant upside during structural stress. On May 10, Romania was managing imbalance throughout the day. A BESS with balancing dispatch capability captured a third revenue stream on top of DAM and ancillary.
What the full annual stack looks like under normal conditions
On days like May 10, the stack compresses into a few hours. Under normal operating conditions across a full year — DAM arbitrage at 30–40 EUR/MWh spread, FCR/aFRR at 60–70 EUR/MWh, balancing market participation at 80–120 EUR/MWh — a 150 MW/300 MWh BESS in Romania generates EUR 14–20 million annually.
May 10 was not the base case. It was the stress case becoming visible.
January 2027: The fourth layer activates
Under GEO 59/2025, Guarantees of Origin (GoO) cross-border certificate trading becomes available to Romanian BESS operators from January 2027. This adds a fourth commercial mechanism — green certificate income sitting on top of the three-layer energy market revenue stack.
Projects that commission in 2026 and are operational in January 2027 capture the full stack from their first operating year. Projects that commission later join a more mature GoO market with more participants and more competition.
The timing is not trivial.
📅 The 2026–2028 Window — Why This Moment Doesn’t Come Back
Unit 2 returns around June 1. Pressure eases. This isn’t the beginning of a permanent crisis.
It’s the preview.
In December 2026, Unit 1 enters refurbishment. Romania runs on 700 MW of nuclear baseload — not 1,400 MW — for two years. During that window, four structural conditions converge:
Condition 1: Lower dispatchable backup. The midday surplus / evening peak differential that produced the 200 EUR/MWh spread in April doesn’t moderate. It widens. One reactor means less dispatchable cushion on days when solar is low and demand is high.
Condition 2: Higher renewable penetration. Solar capacity additions accelerate under AccelerateEU targets. More solar means more midday surplus, more frequent export at low prices, and a deeper evening deficit. The 28 million EUR monthly outflow number doesn’t improve — it grows.
Condition 3: PICASSO aFRR integration matures. Romania’s participation in the European aFRR balancing platform deepens through 2027–2028. Cross-border ancillary market access supports clearing prices in the FCR/aFRR stack and creates new dispatch opportunities for BESS assets pre-qualified under Transelectrica’s ancillary service framework.
Condition 4: GoO trading activates. January 2027.
These four conditions overlap for the duration of Unit 1’s refurbishment — December 2026 through December 2028. Projects that commission during this window operate at the peak convergence of structural grid need and commercial revenue opportunity.
Post-2028, Unit 1 returns. The dispatchable balance improves. The exceptional spread conditions moderate. Projects commissioning after that date operate in a more normalised market.
The entry window is specific. It doesn’t repeat.
🏗️ What Bankable BESS Architecture Looks Like in This Context
How independent engineers are re-reading Romanian revenue projections
When an IE assesses a Romanian BESS project, they model expected DAM spreads under conservative grid conditions — historically 30–40 EUR/MWh. They model FCR/aFRR clearing prices against observed market data. They apply LFP degradation curves (0.7% annually under warranty, approximately 1.0% post-warranty), model augmentation CAPEX at year 12–15, and stress-test DSCR at 1.35× across the full debt tenor — not just year one.
May 10 is now a data point in that picture. An IE reviewing a Romanian BESS project in late 2026 has observed 200 EUR/MWh DAM spreads during a structural grid event — one that will recur systematically during Unit 1’s refurbishment, not as a random shock.
IEs won’t adopt May 10 as the base case. Conservative methodology doesn’t work that way. But it shifts the framing of what qualifies as a stress scenario, what the p90 downside looks like, and what range of spread conditions produces debt service coverage. The 30–40 EUR/MWh base assumption now has a documented outlier at 200 EUR/MWh that occurred because of a structural condition lasting two years.
That changes the credit committee conversation. Not dramatically — but at the margin, and the margin is where deals close or don’t.
The four things lenders actually need to see
1. A multi-mechanism revenue stack. DAM arbitrage, FCR/aFRR, balancing market, GoO from January 2027. Single-mechanism projects — pure arbitrage plays, pure ancillary — face revenue concentration risk questions from lenders. The stack diversifies that risk and demonstrates real grid service capability across operating conditions.
2. A technically credible OT architecture. Network Code on Cybersecurity requirements apply to BESS assets connected to Romania’s transmission network. NIS2 extends those obligations. The EMS must demonstrate operational technology isolation from corporate IT, no persistent remote access channels through Chinese OEM systems, and a cybersecurity posture that survives independent technical review. Independent engineers are asking about this now. Lenders are asking about it now. Preparing these answers at credit committee stage is structurally late.
3. A financial model built around conservative spreads, not May 10. The stress test that matters is not whether the project is viable at 200 EUR/MWh spreads — it obviously is. The test is whether it covers at 1.35× DSCR under 35 EUR/MWh normal conditions, with May 10–type events modelled as periodic upside rather than baseline assumption. Projects that need May 10 to cover their debt service don’t survive IE review.
4. An augmentation plan. LFP chemistry degrades. 0.7% annually under warranty conditions, approximately 1.0% post-warranty. Over a 20-year project life, the effective capacity of the asset shrinks meaningfully without augmentation. Modelling that CAPEX event at year 12–15 is not optional — it’s one of the first things a lender’s technical advisor checks before FID. Projects that omit it get restructured. Sometimes they get restructured after commitment, which is more expensive than doing it correctly in the model.
🎯 What Developers Should Be Doing Before the Window Opens
This section is not strategic. It’s operational.
ATR applications: Transelectrica’s grid connection queue runs two to four years. If you’re targeting a 2027 commissioning date and your ATR process hasn’t started, your commissioning date is aspirational. Start the CTE study now — the ATR agreement is what validates the grid connection, and you need it before lenders will take the technical package seriously.
Revenue model calibration: Review your DAM spread assumptions against the post-May 2026 dataset. Not to build 200 EUR/MWh into the base case. To understand whether your existing p50 and p90 assumptions hold under the Unit 1 refurbishment condition, and whether your DSCR coverage at normalised spreads remains where you think it is.
NIS2 / Network Code technical review: If you haven’t had an OT architecture review against Network Code on Cybersecurity and NIS2 implementation obligations, do it before submitting to lenders. The alternative is receiving IE comments during technical due diligence — at the moment when project timelines are most sensitive to delay.
OEM contract review: The supply chain decisions you make in 2026 — specifically on EMS architecture and the remote access provisions buried in OEM service contracts — determine your cybersecurity posture for the project’s operating life. This is not a procurement question. It is a bankability question. Review the contracts before you sign them, not after.
GoO eligibility: Ensure your development structure and registration process qualifies for GoO cross-border certificate trading under GEO 59/2025. If the corporate structure or project registration doesn’t qualify, January 2027 arrives without the fourth revenue layer. Fixing it retroactively is possible. It is also slower and more expensive than structuring correctly upfront.
The grid stress-tested itself on May 10, 2026.
The results were public, legible, and precise: 140 EUR/MWh spot price, 2,200 MW of imports in a single evening, a 200 EUR/MWh DAM spread that nobody captured because there wasn’t enough storage on the Romanian grid to capture it. Behind that single event: a 28 million EUR monthly outflow that had been running since April, driven not by crisis but by a grid architecture selling cheap and buying expensive every single day.
From December 2026, the conditions that made May 10 visible become the structural baseline.
Battery storage is not the only answer to that problem. It is the fastest deployable answer — 18 to 24 months from FID to commissioning, which puts projects squarely within the Unit 1 refurbishment window if development decisions are made now. The projects that commission in 2026–2027 operate through the peak convergence of grid need and commercial opportunity. The ones that commission in 2029 operate in a different market.
The arithmetic isn’t complicated. The decisions are.
— Tudor, RENEWD